International Journal of Computer & Mathematical Sciences

IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

Copy-Move Image Forgery Detection Using Normalized

Colour Histogram Difference and Scale Invariant Feature
Transform
Rajeev Rajkumar Irengbam Survastava Singh Parhai Moshil Maring
MIT Takyelpat, Imphal MIT Takyelpat, Imphal MIT Takyelpat, Imphal

ABSTRACT
As the world is changing in its Information Technology, it is hard to know the image produced is real or fake.
Many could produce fake images after running several softwares like Photoshops and other powerful tools. With the
images produced from such software, it could change the world or destroy the world, tighten or loosen relationships, etc.
To practically stop from such tamperedimage, we used normalized colour histogram difference and scale invariant
feature transform to detect the tempered part.In this paper, detection of copy-move forgery in any images is done using
normalized colour histogram difference and scale invariant feature transform.After running several processes like
addition of noise, image rotation, etc. we conclude this as the most robust and efficient experiment than others and
charted the colour histogram of the tempered images. We analyse on the standard images provided.
Keywords
Image forgery,scaling,rotation,noise, normalized colour histogram difference,scale invariant feature transformbased
algorithm.

Introduction
Through computer algorithm, digital image processing could be a one of the process among it that
plays a vital role everyday life. Today, due to the availability of large no. of free photo editing software, it is
very easy as well as simple to manipulate the digital images without leaving any traces of tampering. With the
help of tempering process, it interferes the damages or make unauthorized alterations. One could use the
tempered image as proofs in various purposes which results for malicious purposes.

Fig: Different types of Image Forgery

Image forgery is one of the main function comes under the digital image forgery. It studies with techniques or
algorithm to detect traces of digital image tampering. The availability of any of these traces is proof that an
image has been tampered. There are many algorithms or techniques for detecting tampered image. In general,
these techniques can be divided into two major groups: Active Method and Passive Method.
Active Method is a type of method in which certain information is inserted inside an image during formation
or before the image is being spread out to the public.It can be divided into two types – ‘digital watermarking’
and ‘digital signatures’.Watermarking is one of the techniques under Active method. Watermarking requires
special hardware or software in order to fit certain information to the image. Digital watermarking is to be
inserted into the images at the time of recording using special equipment’s this is one of the main
disadvantages of this approaches.
In passive method they do not require any pre-'image distribution' information to be embedded inside digital
image. This method works entirely by inspecting binary information ofDigital image, in the absence of any

56 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

need for external information. In passive method techniques, two methods are divided for the purpose. They
are forgery dependent methods and forgery independent methods. In passive method technique it can be used
to detect traces of image forgery but it can’t guarantee an image that passed the test is original or it has not
been modifies in any manner.’
In digital image forgery, image retouching is one of the less harmful types of digital image forgery comparing
to other types. In image retouching original image does not somewhat changes, but the features of the original
image are enhancement or it reduces. Image Retouching is mostly popular among magazine photo editors they
apply this technique to build up certain features of an image to make it more attractive.

Fig: Image retouching

Image splicing also known as photomontage is a technique use for making forgery where it is more aggressive
type than image retouching. It is a simple type of process and it can be done by cropping and pasting regions
of the same or separate sources. This method is mostly refers to a digital tools available such as in Photoshop.
In this procedure two or more images are combined to create a fake image.

Fig: Image splicing

Copy-move is one of the most famous and commonly use photo tampering method because of its easy to
carry. It is also one of the mostly used technique in creating forgery in digital images. It is a type of pixel
based method. In this method we can copy some of the area in an image and moving the same area to some
content in another area of the image. If the image is measured in different segment, copy move cannot detect
the forged region. In this paper, detection of copy move forgery in any images is done using two methods
called normalized colour histogram difference and scale invariant feature transform.

Original image Tampered image

Fig: Copy-move forgery
Normalised colour histogram is the method that enhances the contrast of images.In this method, changing the
histogram of the original image is done through normalized cumulative sum. The changed images having
original intensity value are then mapped to new intensity to give a uniform histogram of specific intensity
values.A histogram is a graph that shows frequency of anything. Usually blocks in terms of bars represents the
frequency of occurring data in the whole data set.It has two axesnamely X-axis andY-axis.X-axis contains
event whose frequency it should have to count whereas in Y- axis it contains frequency.

57 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

Normalized histogram is one of the methods where the contrast images are enhanced. In this technique,
normalized cumulative sum is converted and use in the histogram of the original image. The intensity values
of the original image are then mapped to a new intensity to produce a uniform histogram of intensity values. It
can occupied various interpolation for this purpose.
Histogram Processing
•Histogram of a gray-level image in the range of [0, L-1] is the discrete function h(rk) =nk, where rkis the
kthgray level and nkis the number of pixels having gray level rk.
•p(rk)can be considered to give an estimate of the probability of occurrence of gray level r k.
•The histogram h(rk) can be normalized to p(rk)=nk/n, for k=0,1,…,L-1.
•The cumulative of all components/probabilities of the normalized histogram is equal to 1.
Histogram of an image shows frequency like others. On the contrary, image frequency of pixels’ intensity
value is shown. In an image histogram, the gray level intensities and the frequency of these intensities are
represented by X-axis and Y-axis respectively.
The histograms of the respective images are given below-

Fig: Histograms of the given image

The algorithm that is utilized to describe the computer vision to detect the image features may be term as
scale-invariant feature transform which is used to the addition to extract key point. Another paramount
characteristic of these features is that the relative positions between them in the pristine scene shouldn't
transmute from one image to another. It can analyse objects even among disorder and under unfinished
blocking, because the SIFT feature descriptor is alike to systematic scaling, orientation, and unfinishedstable
to affine bend and brightness changes.
The key point of the images objects of scale invariant feature transform16 which are stored at the database are
initially derive first from the bulk of reference images. An object of the incipient image is apperceived by
comparing individually one by one the feature from the new image to this database and finds the same
features depending on feature vectors of Euclidean distances.On analysing the steps of the SIFT algorithm,
three steps are involved. They are:
a. Scale-space Extrema Detection:
Scale-space Extrema Detection is a technique where scale space filtering is used for detecting large corners. In
Laplacian of Gaussian it is found that image with various values where LoG acts as a blob detector which
detects blobs in various sizes due to change in . In short, acts as a scaling parameter. But this LoG is a
little costly, so SIFT algorithm uses Difference of Gaussians which is an approximation of LoG. Difference of
Gaussian is obtained as the difference of Gaussian blurring of an image with two different , let it be
and . This process is done for different octaves of the image in Gaussian Pyramid. It is represents in the
image below:

Images are searched for local extrema once DoG are found over scale and space.

58 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

b. Keypoint localization:
Taylor series expansion of scale space is used when keypoints locations are found to get more precise results
and location of extrema, and it rejects the value when the intensity of the extrema is lower than threshold. It is
known as contrast Threshold in OpenCV.
DoG has higher replication for edges, so edges withal need to be abstracted. For this, a concept homogeneous
to Harris corner detector is utilized. It utilizes a 2x2 Hessian matrix (H) to compute the principal curvature.
From Harris corner detector that for edges, one Eigen value is more immensely colossal than the other by
utilizing simple function. In OpenCV it is called edge Threshold if the ratio is more than a threshold and it
discards and eliminates any low-contrast keypoints with edge keypoints and it remains its vigorous interest
points.
c. Orientation Assignment:
In order to achieve invariance to image rotation each keypoint is assign where a neighborhood is taken over
the keypoint location which it depends on the scale, and the gradient magnitude and direction is calculated in
that part. In order to contribute stability of matching an orientation of histogram with 36 bins covering 360
degrees is created with the highest peak in the histogram that is taken and any peak above 80% of it is also
considered to calculate the orientation. It then builds keypoints with equivalent location and scale, but distinct
directions.
d. Keypoint Descriptor:
It is taken when a keypoint descriptor of 16x16 neighborhood is created around a keypoint. It is then divided
into 16 sub-blocks of 4x4 size. For each sub-block, 8 bin orientation histogram is engendered. So a total of
128 bin values are available. It is represented as a vector to compose keypoint descriptor. Several measures
are to be taken to accomplish robustness across illumination changes, rotation etc.
e. Keypoint Matching:
Two images between keypoints are matched by classifying their most proximate neighbors. But in some cases,
the second most proximate-match may be very proximate to the first. It may transpire due to noise or some
other reasons. In that case, ratio of most proximate-distance to second-most proximate distance is taken. If it is
more preponderant than 0.8, they are abnegated. It eliminates of about 90% of false matches which is to be
discards only 5% of correct matches, as per the paper.
In Block-based approach the image for detecting forged area is subdivided into overlapping or non-
overlapping blocks and the characteristics of each block of the image calculated and compared with each
other. All the similar blocks which are identified by lexicographic sorting a matrix of feature vectors is built
so that every feature vector becomes a row in the matrix. Then the matrix is row-wise arrange. Thus, the most
kindred features appear in consecutive rows. Kindred attribute criteria may be Euclidian distance, correlation
etc. The block size affects performance of algorithm. If it is profoundly and immensely vast then it cannot
locate small replicated regions.
Inkey point-based method is to find image Keypoint and collect image features at the detected key points.
Keypoint are locations that carry distinct information of the image content. Each key point is characterized by
a feature vector that consists of a set of image statistics collected at the local neighbourhood of the
corresponding key point.

Literature Survey
In Fridrich AJ Et al1, it chalks out the problems of detecting the copy-move forgery and made a conclusion of
efficient and reliable detection method. Even after the forged part is merge with the background of the image
save in a lossy format, such as JPEG, it could locate the area where it is done.
In Mahdian B Et al2 paper, it works out a method that automatically detect and localize duplicated regions in
digital images. Images based on blur moments are considered. These modifications are commonly used
techniques to conceal traces of copy–move forgery. The method works equally well for lossy format such as
JPEG.
In Bo X Et al3, the paper is based on the SURF (Speed up Robust Features) descriptors, which are invariant to
rotation, scaling etc. where results of experiments indicate that the proposed method is valid in detecting the
image region duplication and quite robust to additive noise and blurring.

59 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

In Huang H Et al4, the paper describes an effective method of detecting Copy-Move forgery in digital images.
In the method, SIFT descriptor extracts the images which have no fix shapes and can be changed in
illumination, rotation, scaling etc. The similarity between pasted region and copied region, the extractor
compares two images to check any possible forged part. The experiment checks the efficiency and quantify its
robustness and sensitivity to post image processing, like additive noise and lossy JPEG compression etc., or
even compound processing.
In Bayram S Et al5, a new approach is considered for detecting copy-move forgery in digital images which is
more robust to lossy compression, scaling and rotation type of manipulations. In the experimental results the
proposed features can detect duplicated region in the images very accurately, even when the copied region
was undergone severe image manipulations and also improved in time efficiency at the expense
In Pan X Et al6 work, it describes a new detection method based on matching image SIFT features. The
robustness of the SIFT features with regards to local transforms renders this method able to detect general
region duplications with efficient computation. The effectiveness of this method is demonstrated with
experimental results, both qualitatively and quantitatively in terms of the detection accuracy and the false
positive rate.
In Hashmi MF Et al7, it developed an algorithm of image-tamper detection based on the Discrete Wavelet
Transform i.e. DWT which is used for dimension reduction, which in turn increases the accuracy of results.
First DWT is applied on a given image to decompose it into four parts LL, LH, HL, and HH. Since LL part
contains most of the information, SIFT is applied on LL part only to extract the key features and find
descriptor vector of these key features and then find similarities between various descriptor vectors to
conclude that the given image is forged. The method allows to detect whether image forgery has occurred or
not and also localizes the forgery i.e. it tells us visually where the copy-move forgery has occurred.
In Yadav P Et al8 paper an algorithm based on Discrete Wavelet Transform (DWT) is used to detect such
cloning forgery. Firstly, DWT is applied to the input image for a reduced dimensional representation then the
compressed image is divided into overlapping blocks. After that Lexicographic sorting is performed, and
duplicated blocks are identified. Due to DWT usage, detection is first carried out on lowest level image
representation. In this technique it increases the accuracy of detection process and reduces the time needed for
the detection process.
In Li Y Et al9, paper a new searching technique is proposed on a polar cosine transform and approximate
nearest neighbour searching based copy-move forgery detection algorithm and it starts by dividing the image
into overlapping patches. Then the potential copy-move pairs are then detected by identifying the patches with
similar features, which is formulated as approximate nearest neighbour searching and accomplished by means
of locality-sensitive hashing (LSH). In this approach it can accurate detection results, and also exhibits high
robustness to various post-processing operations.
In Li L Et al10 paper a method was proposed for detecting a kind of image tampering based on circular pattern
matching. The image is first filtered and divided into circular blocks. A rotation and scaling invariant feature
is then extracted from each block using Polar Harmonic Transform (PHT). The feature vectors are then
lexicographically sorted, and the forged regions are detected by finding the similar block pairs after proper
post-processing. Experimental results demonstrate the efficiency of the method.
In Cozzolino Et al11a new algorithm based on rotation-invariant features is proposed for the accurate detection
and localization of copy-move. A dense-field technique is proposed to overcome the limitation, a fast
approximate nearest-neighbour is restart in order to look upon the algorithm, Patch Match, especially suited
for the computation of dense fields over images. The matching algorithm is adapting to deal efficiently with
invariant features, so as to achieve higher robustness with respect to rotations and scale changes. Moreover,
leveraging on the smoothness of the output field, it implements a simplified and reliable post processing
procedure. The experimental analysis to be at least as accurate, moreover robust, and typically is much faster
than the state-of-the-art dense-field references.
In Chihaoui T Et al12paper, it proposes a method which automatically detects duplicated regions in the same
image. The duplicated detection is performed by identifying the local characteristics of the images (points of
interest) using the Scale Invariant Feature Transform (SIFT) method and by matching between identical
features using the Singular Value Decomposition (SVD) method. Obtained results show that the proposed

60 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

hybrid method is robust to geometrical transformations and is able to detect with high performance duplicated
regions.
In Muhammad N Et al13paper, an efficient non-intrusive method for copy-move forgery detection is presented.
The method is based on image segmentation and similarity detection using dyadic wavelet transform (DyWT).
Copied and pasted regions are structurally similar and the structural similarity is detected using DyWT and
statistical measures. The results show that the proposed method outperforms the stat-of-the-art methods.
In Kakar P Et al14, the paper a novel technique is propose based on transform-invariant features obtained by
using the features from the MPEG-7 image signature tools. In this technique it detects copy-paste forgeries,
along with translation, scaling, rotation, flipping, lossy compression, noise addition and blurring. It also
obtains a high true positive rate and low negative rate with 90% accuracy matching to detect the cloned
regions than the state of the art.
In Lee JC Et al15, this paper a blind forensics approach is propose to detect the input image segmented into
overlapping blocks, whereupon a histogram of orientated gradients where applied to each block. Statistical
features are extracted and reduced to facilitate the measurement of similarity. In these experiment results it
enables to detect multiple examples of copy–move forgery and precisely locate the duplicated regions, even
when dealing with images distorted by translation involving small rotations, blurring, adjustment of radiance,
and colour reduction.

Proposed system
In this paper, two images of original image and tempered image are taken. The two images are divided into
4x4 block size. For each block in two images, normalised colour histogram is computed. Depending upon the
threshold, if the similarities between the original image and tampered image are less than zero (0), than they
are treated as unequal, otherwise it is equal. The unequal blocks are selected and SIFT (scale invariant feature
transform) is applied to the selected area. Than the selected blocks are tampered and detected in the tampered
image. The selected blocks are highlighted with green colour.
Input: Original Image (I) and tampered image (J).
Output: Tampered region in J.
Procedurecopy_move_detection (I, J)
I  read the original image
J  read the tampered image
B1  divide I into blocks of size b  b and arrangein order
B2  divide J into blocks of size b  b and arrangein order
n  number of blocks
for x  1 to n  1
for y  x  1:n
HD  apply color histogram difference between B1 x and B2 y
if HD  T //where T is a threshold
S1  apply SIFT in B1x
S 2  apply SIFT in B2y
M  match feature between S1 andS2
Display the matched region if number of match points in M is 1 or more
else
SIFT det ection is not applied
end if
end for
end for

61 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

Implementation and Discussion

The images after experimenting using normalized colour histogram difference and scale invariant feature
transform, the following figures gives the tempered images of their respective images.

A1. Original A2.Tempered A3. Detected Area

Fig1. A2 is the corresponding tampered image of the original image A1.

B1. Original B2. Tempered B3. Detected area

Fig2. B2 is the tampered image of the corresponding original image B1.

C1. Original C2. Tampered C3. Detected area

Fig3. C2 is the tampered image of the corresponding original image C1.
Conclusion & future work
With all these experimental results, it may be concluded as the most stable and effective in detecting the
tempered image of the specific region. Using normalized colour histogram difference and scale invariant
feature transform, tempered image can be detected easily. Further, more areas can be broaden to avail certain
benefits like on military aspects. With these issues, advantages on social crisis, developmental works can be
updated and widen up.

62 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring

 International Journal of Computer & Mathematical Sciences
IJCMS
ISSN 2347 – 8527
Volume 6, Issue 3
March 2017

References
[1] Fridrich AJ, Soukal BD, Lukáš AJ. Detection of copy-move forgery in digital images. InProceedings of Digital
Forensic Research Workshop 2003.
[2] Mahdian B, Saic S. Detection of copy–move forgery using a method based on blur moment invariants. Forensic
science international. 2007 Sep 13;171(2):180-9.
[3] Bo X, Junwen W, Guangjie L, Yuewei D. Image copy-move forgery detection based on SURF. InMultimedia
information networking and security (MINES), 2010 international conference on 2010 Nov 4 (pp. 889-892).
IEEE.
[4] Huang H, Guo W, Zhang Y. Detection of copy-move forgery in digital images using SIFT algorithm.
InComputational Intelligence and Industrial Application, 2008. PACIIA'08. Pacific-Asia Workshop on 2008
Dec 19 (Vol. 2, pp. 272-276). IEEE.
[5] Bayram S, Sencar HT, Memon N. A survey of copy-move forgery detection techniques. InIEEE Western New
York Image Processing Workshop 2008 Sep (pp. 538-542). IEEE.
[6] Pan X, Lyu S. Detecting image region duplication using SIFT features. InAcoustics Speech and Signal
Processing (ICASSP), 2010 IEEE International Conference on 2010 Mar 14 (pp. 1706-1709). IEEE.
[7] Hashmi MF, Hambarde AR, Keskar AG. Copy move forgery detection using DWT and SIFT features.
InIntelligent Systems Design and Applications (ISDA), 2013 13th International Conference on 2013 Dec 8 (pp.
188-193). IEEE.
[8] Yadav P, Rathore Y. Detection of copy-move forgery of images using discrete wavelet transform. International
Journal on Computer Science and Engineering. 2012 Apr 1;4(4):565.
[9] Li Y. Image copy-move forgery detection based on polar cosine transform and approximate nearest neighbour
searching. Forensic science international. 2013 Jan 10;224(1):59-67.
[10] Li L, Li S, Zhu H, Wu X. Detecting copy-move forgery under affine transforms for image forensics. Computers
& Electrical Engineering. 2014 Aug 31;40(6):1951-62.
[11] Cozzolino D, Poggi G, Verdoliva L. Efficient dense-field copy–move forgery detection. IEEE Transactions on
Information Forensics and Security. 2015 Nov;10(11):2284-97.
[12] Chihaoui T, Bourouis S, Hamrouni K. Copy-move image forgery detection based on SIFT descriptors and SVD-
matching. InAdvanced Technologies for Signal and Image Processing (ATSIP), 2014 1st International
Conference on 2014 Mar 17 (pp. 125-129). IEEE.
[13] Muhammad N, Hussain M, Muhammad G, Bebis G. Copy-move forgery detection using dyadic wavelet
transform. InComputer Graphics, Imaging and Visualization (CGIV), 2011 Eighth International Conference on
2011 Aug 17 (pp. 103-108). IEEE.
[14] Kakar P, Sudha N. Exposing postprocessed copy–paste forgeries through transform-invariant features. IEEE
Transactions on Information Forensics and Security. 2012 Jun;7(3):1018-28
[15] Lee JC, Chang CP, Chen WK. Detection of copy–move image forgery using histogram of orientated gradients.
Information Sciences. 2015 Nov 10; 321:250-62
[16] Naveen C. SIFT (Scale invariant Feature Transform) Algorithm.

63 Rajeev Rajkumar, Irengbam Survastava Singh, Parhai Moshil Maring